Read the schema before the screen
Labels are a story. Fields are a contract. Audits go faster when you start with the names the system already uses.
A screen that says ‘reference number’ may post ‘legacy_acct_cd’. An export that says ‘invoice date’ may mean ‘posted date’. Customers do not see that gap until a payment lands on the wrong month.
We ask for a small sample of contracts at the start of a schema check: a payload, a CSV, a data dictionary if one exists. Then we walk the screens with those names in hand. Disagreement is the finding. The UI is only the place it becomes visible.
Teams sometimes worry this is too engineering-heavy for a customer-facing review. It is the opposite. Customers experience schema drift as broken promises: the wrong amount, the missing document, the profile that will not save.
You do not need a perfect model. You need the current one, written down, so the next release cannot quietly rename a field that statements still depend on.